This project is a very big part of my Master Thesis at WNE UW, you can expect:

• data preprocessing (missing detection and imputation with earlier developed RF based algorithm, data manipulation, image to vector conversion using Monte Carlo simulation)
• data analysis and transforms (balance check, distributions analysis (box plots and histograms), outliers reduction using quantile clipping, Yeo-Johnson power transform, normalization, special normalization variant for CNNs)
• n-Fold Cross Validation study on training set and testing on test set
• several algorithms (Logits, Random Forests, XGBoosts, Feedforward Neural Networks, Convolutional Neural Networks)
• hyperparameters optimization for Logits (regularization only), RFs and XGBs
• Networks building (experiments based on results and learning curves)
• optimized Inception modules for CNNs
• automatic feature generation method for enlarging CNNs inputs, based on composing sampled features and arithmetic operations from created discrete probability distributions many times
• CV results comparison (Wilcoxon testing)
• models training on the entire training set (specifications chosen in CV)
• models testing and comparison
• study on 5 datasets - companies bankruptcy prediction for 1-5 years forecast horizons (in progress)
• Python notebooks in English (a ton of Python code)

• Computer Vision
• finance
• Machine Learning
• Neural Networks
• notebook
• Python
• statistics
• Thesis